Big 5 AI Vendor Roundup: Week of May 18, 2026

Research By: Mark Tauschek, Bill Wong, Info-Tech Research Group

Google I/O dominated the week with Gemini 3.5 Flash, Gemini Omni, and Gemini Spark rolling out across the entire Google product portfolio. Anthropic hired OpenAI co-founder Andrej Karpathy, acquired Stainless, landed strategic alliances with KPMG and PwC, published a Mythos vulnerability update, and is preparing to brief the Financial Stability Board on what Mythos has been finding. The Musk v. Altman trial ended with a unanimous jury verdict dismissing all claims on statute of limitations grounds, clearing the path for OpenAI’s confidential IPO filing, which reportedly followed within 48 hours. SpaceX filed its S-1 the same day. Nvidia posted record Q1 FY2027 revenue of $81.6 billion. AWS shipped multicloud connectivity with Oracle and marked the one-year anniversary of AWS Transform. OpenAI and Google both announced they are expanding SynthID watermarking and C2PA content credentials across their ecosystems.

The model wars have a new entrant (Gemini 3.5 Flash), and the IPO wars have started. This is the week the public markets entered the AI conversation in a way they haven’t before.

Google: I/O 2026 Was the Main Event

Google I/O delivered 100 announcements across two days. Here is what matters for enterprise buyers and IT leaders.

  • Gemini 3.5 Flash. The headline model announcement. Google says 3.5 Flash surpasses the prior 3.1 Pro in coding, agentic, and multimodal benchmarks while maintaining Flash-series cost and speed, at 4x faster output tokens per second than other frontier models. It is rolling out now in the Gemini app, Search, Antigravity 2.0, and the Gemini API. Gemini 3.5 Pro is in testing and expected next month. For enterprises building on the Gemini API or Agent Platform, 3.5 Flash is the model to evaluate immediately.
  • Gemini Omni. A new model series that combines reasoning with creation. Omni accepts image, audio, video, and text input and outputs video grounded in real-world knowledge. The editing capability is conversational, meaning you can refine generated video with natural language. Omni Flash will roll out to developers and enterprise customers via the Gemini API and Agent Platform API in coming weeks.
  • Gemini Spark. This 24/7 personal AI agent that autonomously takes action on your behalf is available for Gemini Enterprise and Workspace customers. This is the productized version of the (codenamed) "Remy" agentic assistant that was previewed ahead of I/O, now positioned as the core Workspace assistant going forward. Includes a Daily Brief feature and persistent task management.
  • Universal Cart. Google's agentic commerce play. An intelligent shopping cart that tracks deals and works across apps and websites. Combined with the Information Agents in Search (background 24/7 monitoring agents for AI Pro and Ultra subscribers), Google is positioning Gemini as the layer between users and the entire web.
  • Search upgraded to Gemini 3.5 Flash. AI Mode, which Google says now has more than 1 billion monthly users, is upgraded to 3.5 Flash as the default model globally. Google also launched what it describes as the biggest upgrade to the Search box in over 25 years, with AI-powered query suggestions that anticipate intent.
  • Google Antigravity 2.0. The agent-first development platform gets expanded capabilities and new integration with Agent Platform, bringing agentic development to enterprise customers at scale.
  • Intelligent Eyewear. Samsung smart glasses with Gemini built in, launching this fall. Consumer-focused, but the form factor signals where enterprise AR use cases are heading.
  • The enterprise takeaway. Google just shipped a new model family (3.5), a new multimodal model (Omni), a 24/7 agent for Workspace (Spark), and a refreshed developer platform (Antigravity 2.0) in a single week. Combined with last week’s Gemini Intelligence for Android, Google has the broadest AI surface of any vendor: phone, laptop, browser, workspace, cloud, search, and now glasses. The challenge for enterprise buyers is evaluation fatigue. There is a lot to assess, and much of it is rolling out, so not everything is generally available today. Wait for GA dates before committing.

Anthropic: Karpathy Hire, Stainless Acquisition, Mythos Update, Professional Services Land Grab

Anthropic made four different kinds of moves this week: talent, infrastructure, cybersecurity, and distribution.

  • Andrej Karpathy joins Anthropic (May 19). The OpenAI co-founder and former Tesla AI director announced he is joining Anthropic’s pretraining team under team lead Nick Joseph. Karpathy will build a new team focused on using Claude itself to accelerate pretraining research, which is a direct investment in recursive model improvement. “I think the next few years at the frontier of LLMs will be especially formative,” Karpathy wrote. This is the highest-profile talent move in the AI industry this year. Karpathy co-founded OpenAI, led Tesla’s Autopilot AI, returned to OpenAI, left again in 2024, and is now at Anthropic. The hire follows Ross Nordeen (founding xAI member) joining Anthropic earlier this month. Anthropic is systematically pulling talent from every major AI lab.
  • Stainless acquisition (May 18). Anthropic acquired Stainless, the startup behind every official Anthropic SDK since the API launched. Stainless generates SDKs, CLIs, and MCP servers from API specifications. Hundreds of companies, including OpenAI, Google, and Meta, use Stainless tooling. The acquisition gives Anthropic direct control over the developer experience layer and, critically, over MCP server generation as agentic AI moves into production. The battle has shifted from pure model performance to system integration and developer tooling, and this acquisition reflects that directly.
  • Mythos vulnerability update and FSB briefing. Two Mythos stories broke this week. First, Anthropic published a Project Glasswing update reporting that Mythos Preview has now found over 23,000 total vulnerabilities across more than 1,000 open-source projects, of which it estimates 6,202 are high or critical severity. Of the 1,900 findings triaged so far, external security research firms have confirmed a 90.8% true positive rate. As of May 22, 1,596 vulnerabilities have been disclosed to maintainers across 281 projects. Partners are reporting a tenfold increase in their rate of bug-finding, with Cloudflare alone discovering 2,000 bugs including 400 rated high or critical. Microsoft’s recent acknowledgment that its patch releases will “continue trending larger for some time” is reportedly linked to bugs found through Mythos Preview. Second, the Financial Times reported that Anthropic will brief the Financial Stability Board (FSB) on cyber vulnerabilities Mythos has identified in the global financial system. The briefing was requested by Bank of England governor and FSB chair Andrew Bailey, who said publicly last month that Anthropic “may have found a way to crack the whole cyber risk world open.” Only about 50 organizations currently have Mythos access, including Amazon, Microsoft, and JPMorgan Chase. The FSB briefing is the first time a frontier AI lab is presenting vulnerability findings directly to G20 finance ministries and central banks.
  • KPMG global alliance (May 19). KPMG announced it is embedding Claude into Digital Gateway, the platform KPMG’s people and clients use for delivery, starting with tax and private equity. All 276,000+ KPMG employees globally gain access to Claude. Anthropic named KPMG a preferred partner for private equity, and the two companies will codevelop Claude-powered products for PE portfolio companies.
  • PwC expanded partnership (May 14, rolled out this week). PwC is deploying Claude Code and Cowork starting with US teams and expanding globally. The two organizations are establishing a joint Center of Excellence and a program to train and certify 30,000 PwC professionals on Claude.
  • The professional services pattern. With KPMG, PwC, and the Blackstone/Goldman/H&F joint venture (JV) from three weeks ago, Anthropic now has distribution partnerships covering three of the Big Four accounting firms plus three of the largest private equity (PE) shops. This is the forward-deployed engineering model at scale, and it is a deliberate strategy to embed Claude into enterprise workflows through the consultants and advisors who build them.
  • Anthropic in talks to use Microsoft Maia chips (May 21). The Information reported that Anthropic is in discussions to rent Microsoft's custom Maia AI chips, adding yet another silicon option to its already diversified compute stack. Anthropic is now running or exploring capacity on Google TPUs, AWS Trainium, Nvidia GPUs (via SpaceX Colossus, Azure, and Akamai), and potentially Microsoft Maia. No other AI lab has this breadth of silicon diversification.

OpenAI: Musk Trial Verdict, Confidential IPO Filing, SpaceX S-1

The legal overhang lifted, and OpenAI moved fast.

  • Jury dismisses all claims in Musk v. Altman (May 18). A federal jury in Oakland unanimously found that Musk waited too long to file his lawsuit, dismissing all claims on statute of limitations grounds after less than two hours of deliberation. The verdict means Altman, Brockman, and OpenAI are not liable on any count. Microsoft was also cleared of aiding and abetting. Musk’s attorney Marc Toberoff said, “I can sum it up in one word: appeal.” OpenAI’s lead lawyer William Savitt called it “a hypocritical attempt to sabotage a competitor.” The verdict does not address the merits of Musk’s breach of charitable trust claims, but it removes the most immediate threat to OpenAI’s for-profit conversion and, by extension, its IPO.
  • OpenAI preparing confidential IPO filing (May 20). Axios, CNBC, and the Wall Street Journal reported that OpenAI is working on a confidential IPO prospectus that could be filed imminently, targeting a public listing as early as September 2026 at a valuation above $1 trillion. Goldman Sachs and Morgan Stanley are advising. The timing is deliberate given the filing reportedly came within hours of SpaceX publishing its S-1. Axios noted the timing “is clearly designed to steal the spotlight from Elon Musk’s upcoming SpaceX IPO.”
  • SpaceX S-1 filed (May 20). SpaceX publicly filed its IPO prospectus on the same day, disclosing $18.7 billion in 2025 revenue (including the xAI merger) and seeking to raise up to $80 billion at a valuation of approximately $1.75 trillion. The filing revealed the Anthropic Colossus deal is worth $1.25 billion per month through May 2029, a detail not previously disclosed. Musk controls 85% of voting power through Class B shares. Starlink accounts for more than two-thirds of revenue.
  • For enterprise buyers, the implication is straightforward: OpenAI’s corporate structure is about to become more transparent than it has ever been. The S-1, when public, will be the first time buyers can see audited financials. Members negotiating OpenAI enterprise agreements should wait for the public filing before finalizing multiyear commitments.
  • OpenAI and Dell Technologies partner to bring Codex to hybrid and on-premises enterprise environments.

Microsoft: Quieter Week, Interconnect Progress

Microsoft had a quieter week as the Musk trial verdict and Google I/O grabbed most of the industry’s attention.

  • AWS Interconnect multicloud connectivity with Oracle (preview). This is technically an AWS announcement, but the multicloud story matters for Microsoft because AWS Interconnect already supports Google Cloud (GA), OCI (preview), and Microsoft Azure (coming later in 2026). When Azure support goes live, enterprises will have private, managed connectivity across all three hyperscalers through a single AWS-managed interface. Microsoft’s own AI startup acquisition discussions (Inception, covered last week) remain ongoing with no resolution reported.
  • Copilot Studio and Agent 365 rollout continues. No major new announcements, but the M365 E7 bundle ($99/user/month) is now generally available. Members should be running comparative evaluations against Claude-on-M365, which is live with Excel, PowerPoint, and Word add-ins, and Outlook in beta.

Amazon: Transform Anniversary, Multicloud Connectivity, Claude Platform on AWS

AWS had a steady infrastructure week.

  • AWS Transform one-year anniversary. The agentic AI service for modernizing enterprise applications now has agents available in Kiro, Claude, Cursor, and Codex, with a new agent builder toolkit for building custom transformation agents. AWS says Transform has saved customers over 1.6 million hours of manual modernization effort.
  • AWS Interconnect multicloud connectivity with OCI (preview). Private, managed connections to Oracle Cloud Infrastructure, joining Google Cloud (GA) and Microsoft Azure (coming later in 2026). This is the infrastructure play for enterprises running agents across multiple clouds.
  • Claude Platform on AWS. Noted in this week’s AWS roundup alongside Bedrock updates. Anthropic’s models continue to deepen integration with AWS infrastructure while Anthropic simultaneously expands compute relationships with every other major provider.
  • Build on Trainium program. AWS invested $110 million to give university researchers access to purpose-built AI chips. UC Berkeley, MIT, Carnegie Mellon, and other institutions are using Trainium to accelerate AI research. This is the long-term talent pipeline investment that keeps AWS competitive in custom silicon.

Bonus: Nvidia Q1 FY2027 Earnings – Record Revenue, Record Dividend, Muted Stock Reaction

Nvidia reported Q1 FY2027 results on May 20, and the numbers speak for themselves.

  • Record revenue of $81.6 billion, up 85% year over year and 20% sequentially, beating the Street consensus of approximately $78.8 billion. Data Center revenue hit a record $75.2 billion, up 92% year over year, driven by the Blackwell 300 ramp and a surge in networking revenue.
  • Data Center networking revenue of $14.8 billion, up 199% year over year. InfiniBand, Spectrum-X Ethernet, and NVLink are generating real revenue as networking is now 20% of Data Center revenue.
  • Nvidia introduced a new reporting framework this quarter as they divided their core Data Center revenue ($75.25 billion) into two distinct segments. The first is based on the hyperscalers ($37.87 billion from Microsoft, Google, Amazon, Meta), and the second is based on ACIE ($37.38 billion, AI Clouds, Industrial, and Enterprise – Sovereign AI is part of this segment and CFO Colette Kress stated that sovereign-driven revenue increased by more than 80% year over year). It should be noted that just six months ago, revenue from the hyperscalers was well over 50% of data center revenue, now the hyperscalers vs. ACIE split is roughly 50/50.
  • Q2 guidance of $91 billion (midpoint), above the Street’s expectation of roughly $86 billion. Nvidia is not assuming any Data Center compute revenue from China.
  • Dividend increased 25x, from $0.01 to $0.25 per share quarterly. Plus, an additional $80 billion share repurchase authorization. Nvidia generated $50.3 billion in operating cash flow in the quarter.
  • Jensen Huang: “Demand has gone parabolic. The reason is simple: Agentic AI has arrived.” Nvidia is the only platform that runs every frontier AI model, Huang noted, naming Anthropic, OpenAI, SpaceX/xAI, Meta, and Google’s Gemini.
  • The stock fell approximately 1.5% after hours. In what has become a pattern, Nvidia has beaten estimates in 18 of its last 20 quarters, and the stock has fallen after the last four reports. The bar is perpetually above the already record results. For enterprise budget planning, the underlying message is that AI infrastructure demand is accelerating, not plateauing, and the $725 billion in combined hyperscaler CapEx for 2026 is being absorbed. The costs to operate AI will not be coming down, and the costs will eventually be passed on to enterprise buyers.

Our Take

Three things happened this week that will define the next 12 months. First, the Musk v. Altman verdict cleared the runway for OpenAI’s IPO, which means enterprise buyers will soon have audited financials for the first time. That changes the conversation from trust to verification. Second, Google I/O shipped a new model family (3.5), a multimodal creation model (Omni), and a 24/7 Workspace agent (Spark) in a single event. Google now has the broadest AI surface of any vendor, and if the evaluation burden on IT leaders wasn’t already overwhelming enough, it just got heavier. Third, Nvidia’s $81.6 billion quarter with Q2 guidance of $91 billion confirms that the CapEx cycle is still accelerating, which means the infrastructure costs underwriting the AI race are going up, and those costs will flow through to enterprise pricing.

The Mythos story continues to evolve and it’s still early days. A 90.8% true positive rate across nearly 2,000 triaged findings, 1,596 vulnerabilities disclosed to open-source maintainers, and a briefing to the Financial Stability Board requested by the chair himself show that this is no longer a product announcement. It is a structural shift in how vulnerabilities are discovered and how regulators are responding. Members should be tracking Glasswing’s public disclosure dashboard and asking their own vendors whether they have Mythos access and what they are doing with the findings.

The talent story seems like a footnote with all the other news, but it may be much more consequential in the near term. With Karpathy moving from OpenAI to Anthropic, Nordeen from xAI to Anthropic, and Microsoft shopping for Inception, the talent and model-ownership landscapes are shifting. The vendor you signed with last quarter may have a noticeably different research team next quarter. Ask who is building the models, not just what the benchmarks say.

Want to Know More?

Latest Research

All Research